Hi,
I think that to avoid ICE-ing during error recovery when DECL_SAVED_TREE
is NULL_TREE we want to simply bail out and return true to the caller.
Tested x86_64-linux.
Thanks,
Paolo.
//////////////////
/cp
2014-12-12 Paolo Carlini <paolo.carl...@oracle.com>
PR c++/59628
* semantics.c (finish_omp_reduction_clause): Early return true
if DECL_SAVED_TREE (id) is NULL_TREE.
/testsuite
2014-12-12 Paolo Carlini <paolo.carl...@oracle.com>
PR c++/59628
* g++.dg/gomp/pr59628.C: New.
Index: cp/semantics.c
===================================================================
--- cp/semantics.c (revision 218683)
+++ cp/semantics.c (working copy)
@@ -5138,6 +5138,8 @@ finish_omp_reduction_clause (tree c, bool *need_de
id = OVL_CURRENT (id);
mark_used (id);
tree body = DECL_SAVED_TREE (id);
+ if (!body)
+ return true;
if (TREE_CODE (body) == STATEMENT_LIST)
{
tree_stmt_iterator tsi;
Index: testsuite/g++.dg/gomp/pr59628.C
===================================================================
--- testsuite/g++.dg/gomp/pr59628.C (revision 0)
+++ testsuite/g++.dg/gomp/pr59628.C (working copy)
@@ -0,0 +1,13 @@
+// PR c++/59628
+// { dg-do compile }
+// { dg-options "-fopenmp" }
+
+struct A { int i; };
+
+void foo()
+{
+ A a;
+ #pragma omp declare reduction (+: A: omp_out.i +: omp_in.i) // { dg-error
"expected" }
+ #pragma omp parallel reduction (+: a)
+ ;
+}